Convolutional Neural Networks Performance Study for Image Processing of Waste Segregation for Reverse Vending Machine

The main element for reverse vending machine (RVM) is the waste classification technique. The performance of the RVM’s waste segregation can be enhanced by incorporating a good classification technique. The objective of this project is to study the performance of CNN for image processing for the waste segregation part of RVM. For this RVM, only polyethylene terephthalate (PET) bottles, aluminum cans, and drink carton boxes are considered for recycling by using image classification based on transfer learning with Convolutional Neural Network (CNN) algorithms. The performance parameters that are evaluated are F1-score, computational time, and testing result for each neural network. In this paper, ResNet50 surpasses other neural networks due to the highest F1-score which is 0.9541 and the good testing performance although the computational time is longest among the other network which is 240 min 01 s. The accuracy rate of ResNet50 achieved in this project is 0.9724.